BACKGROUND: Various bleeding risk scores have been proposed to assess the risk of bleeding in patients with atrial fibrillation taking oral anticoagulants. Limited data are available with these scores, in users of non-vitamin K antagonist oral anticoagulants.
METHODS: Using the Danish registries, we evaluated and compared the risk classification properties of the Hypertension, Age, Stroke, Bleeding tendency/predisposition, Labile international normalized ratios, Elderly age/frailty, Drugs such as concomitant aspirin/nonsteroidal anti-inflammatory drugs or alcohol excess (HAS-BLED), Anticoagulation and Risk Factors in Atrial Fibrillation (ATRIA), and Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT) scores for predicting major bleeding in 57,930 atrial fibrillation patients (44.6% female; mean age 73.5 years, standard deviation 11.4 years; mean CHA2DS2-VASc score 3.2, standard deviation 1.8).
RESULTS: At 1-year follow-up, C-statistics for ATRIA, HAS-BLED, and ORBIT were approximately 0.59, with only minor differences between scores. Both ATRIA and ORBIT categorized more patients as “low risk” (both >83%, when compared with HAS-BLED, only 53%), and qualitatively, the receiver operating characteristic curves revealed higher sensitivity (62.8%) for HAS-BLED compared with ATRIA (29.7%) and ORBIT (37.1%). The clinical usefulness of scores was evaluated using decision curve analyses at a 1-year perspective. If the intervention threshold is low (<1.7%), the benefit is toward monitoring all patients. If preference is for a major bleeding risk threshold between 1.7% and 2.0%, most benefit was obtained by using HAS-BLED. The ORBIT and ATRIA scores provided better benefit for thresholds between 2.0% and 6.0%.
CONCLUSIONS: This analysis of contemporary bleeding risk score stratification in a “real-world” non-vitamin K antagonist oral anticoagulant users population with atrial fibrillation showed modest predictive values using C-statistics. The scores represent different risk thresholds, with HAS-BLED classifying least patients at low risk and achieving the highest benefit if applying a major bleeding intervention threshold of approximately 2%, whereas benefit from using either ATRIA score or ORBIT score was only evident using higher intervention thresholds.